ON THE MAXIMAL GRADED SHIFTS OF IDEALS AND

MODULES
arXiv:1801.08194v1 [math.AC] 24 Jan 2018

JASON MCCULLOUGH

Dedicated to Professor Craig Huneke on the occasion of his sixty-fifth birthday.

Abstract. We generalize a result of Eisenbud-Huneke-Ulrich on the

maximal graded shifts of a module with prescribed annihilator and prove
a linear regularity bound for ideals in a polynomial ring depending only
on the first p − c steps in the resolution, where p = pd(S/I) and c =
codim(I).

1. Introduction
Let S = k[x1 , . . . , xn ] be a polynomial ring over a field k, viewed as a
standard graded ring, and let M be a finitely generated graded S-module.
Let F• be the minimal graded free resolution of M over S, so that Fi =
⊕j S(−j)βij (M ) . Following the notation in [11], we denote the ith maximal
and minimal graded shifts as
Ti (M ) = max{j | TorSi (M, k)j 6= 0} = max{j | βij (M ) 6= 0},
ti (M ) = min{j | TorSi (M, k)j 6= 0} = min{j | βij (M ) 6= 0}.
Other authors [2, 14, 17] have used ti (M ) to denote the maximal shifts but
the above notation seems more apt to the situation of dealing with both
maximal and minimal shifts. The maximal shifts are of primary interest
due to their connection with Castelnuovo-Mumford regularity, that is
reg(M ) = max {Ti (M ) − i}.
0≤i≤pd(M )

Fix M = S/I to be a cyclic module, where I is a homogeneous ideal of

S. A motivating question for this paper, which has received a lot of recent
interest (see e.g. [1], [2], [11], [14]) is the following:
Question 1.1. Which sequences of nonnegative integers occur as {Ti (S/I)}
for some homogeneous ideal I in a polynomial ring S?
There is a doubly exponential upper bound on reg(S/I) (and hence all of
the Ti (S/I)) in terms of T1 (S/I), the maximal degree of a minimal gener-
ator of I. (See [4] and [7].) Examples derived from the Mayr-Meyer ideals
[15] show that this upper bound is nearly optimal in that families of ideals
generated by quadrics in n variables can have first syzygies whose degree
1
2 J. MCCULLOUGH

grows doubly exponentially in n. Other examples of ideals with large regu-

larity (e.g. [3], [4], [5]) also have large degree first syzygies. Later examples
of “designer ideals” constructed by Ullery [19] showed that for any strictly
increasing sequence of positive integers 2 ≤ a1 < a2 < · · · < am , there is an
ideal I in a polynomial ring S such that Ti (S/I) = ai for all 1 ≤ i ≤ m.
However, these ideals also satisfy Tm+i (S/I) = am + i for i > 0 (i.e. there
is a linear strand at the end of the resolution) and the resolutions tend to
be rather long. These examples motivate the idea that ideals whose resolu-
tions show large degree jumps early in the resolution must be “paid for” by
smaller degree jumps later in the resolution. The first result quantifying this
idea is the following weak convexity result due to Eisenbud-Huneke-Ulrich:
Theorem 1.2 (Eisenbud-Huneke-Ulrich [9, Corollary 4.1]). Let I be a ho-
mogeneous ideal in S = k[x1 , . . . , xn ] such that dim(S/I) ≤ 1. Then
Tn (S/I) ≤ Ti (S/I) + Tn−i (S/I),
for all 0 ≤ i ≤ n.
Note that by replacing n with p := pd(S/I), the statement extends to ideals
I such that dim(S/I) − depth(S/I) ≤ 1, since we may always assume k
is infinite and then reduce both S and S/I by a sequence of linear forms
regular on S/I leaving us in the case when pd(S/I) = n. In particular,
Tp (S/I) ≤ Ti (S/I) + Tp−i (S/I) for all 0 ≤ i ≤ p = pd(S/I) when S/I is
Cohen-Macaulay. Previously the author asked [17, Question 5.1] whether the
same weak convexity inequality held without the assumptions on depth(S/I)
or dim(S/I). The author knows of no counterexamples to this statement.
Note also that such a weak convexity inequality would imply a linear bound
on Tp (S/I) for any ideal in terms of T1 (S/I), . . . , Th (S/I), where h := 2p .
 

In [17, Theorem 4.7] the author proved a weaker polynomial bound on

reg(S/I), and in particular on Tp (S/I), in terms of T1 (S/I), . . . , Th (S/I)
without any restriction on depth(S/I) or dim(S/I) by means of the Boij-
Söderberg decomposition of the Betti table of S/I.
Instead of discussing cyclic modules, one can more generally consider
modules whose annihilator contains some fixed ideal. In the same paper as
above, Eisenbud-Huneke-Ulrich proved the following:
Theorem 1.3 (Eisenbud-Huneke-Ulrich [9, Corollary 4.2]). Let M be a
finitely-generated graded S-module of codimension c, projective dimension p,
and with dim(M ) − depth(M ) ≤ 1. Let J be a homogeneous ideal contained
in Ann(M ). If depth(S/J) ≥ depth(M ), then for 0 ≤ q ≤ codim(J),
Tp (M ) ≤ Tp−q (M ) + Tq (S/J).
In particular:
(a) If Ann(M ) contains a regular sequence of degrees d1 , . . . , dq , then
q
X
Tp (M ) ≤ Tp−q (M ) + di .
i=1
 ON THE MAXIMAL GRADED SHIFTS OF IDEALS AND MODULES 3

(b) If J is generated in degree d with linear resolution, then

Tp (M ) ≤ Tp−q (M ) + d + q − 1.
We conjecture that Theorem 1.3 also holds without the almost Cohen-
Macaulay hypothesis.
Conjecture 1.4. Let M be a finitely-generated graded S-module of codi-
mension c. Let J be a homogeneous ideal contained in Ann(M ) and set
p = pd(M ). If depth(S/J) ≥ depth(M ), then for 0 ≤ q ≤ codim(J),
Tp (M ) ≤ Tp−q (M ) + Tq (S/J).
Note that if Conjecture 1.4 is true, then for any homogeneous ideal I with
pd(S/I) ≤ 2 codim(I), Question 5.1 in [17] has an affirmative answer.
The purpose of this note is to give evidence in support of this conjecture
when S/J is Cohen-Macaulay. Note that if true, the conjecture implies
various linear bounds on Tp (M ) in terms of T0 (M ), T1 (M ), . . . , Tp−c (M )
and T1 (S/J), . . . , Tc (S/J).
As the main result of this note, we prove in Theorem 3.5 a weaker linear
bound on reg(M ), and thus also Tp (M ), in terms of the same data without
restriction on depth(M ) or dim(M ). In particular, when M = S/I is a
cyclic module, we get the following regularity bound in terms of only the
first pd(S/I) − codim(I) syzygy degrees.
Theorem (Corollary 3.7). Let I be an ideal of S with p = pd(S/I) and
c = codim(I). Then
reg(S/I) ≤ max {Ti (S/I) + (p − i) T1 (S/I)} − p.
0≤i≤p−c

By adapting an argument of Herzog-Srinivasas [14], we also prove in The-

orem 3.1 the special case of Conjecture 1.4 when J is principal. This recovers
and strengthens results of Herzog-Srinivasan [14] and El Khoury-Srinivasan
[11] as well as the author [17].
In the next section we gather some preliminary results. Section 3 contains
the main results and an example.

2. Preliminary Results
We begin with some elementary facts about the maximal graded shifts in
short exact sequences.
Lemma 2.1. Let 0 → M ′ → M → M ′′ → 0 be a short exact sequence of
graded S-modules. Then for i ≥ 0,
(1) Ti (M ) ≤ max {Ti (M ′ ), Ti (M ′′ )},
(2) Ti (M ′ ) ≤ max {Ti (M ), Ti+1 (M ′′ )},
(3) Ti (M ′′ ) ≤ max {Ti (M ), Ti−1 (M ′ )}.
Proof. All of the statements follow from the long exact sequence of Tor
stemming from tensoring the above short exact sequence with k. 
4 J. MCCULLOUGH

The following statement about the early maximal graded shifts seems to
be well-known but we include a proof for completeness. A similar result is
proved in [12, Lemma 1.2].
Proposition 2.2. If M be a finitely-generated graded S-module of codimen-
sion c, then Ti (M ) < Ti+1 (M ) for 0 ≤ i < c.
Proof. Suppose to the contrary that Ti (M ) ≥ Ti+1 (M ) with 0 ≤ i < c.
Let F• be the minimal graded free resolution of M with differential maps
∂i : Fi → Fi−1 . Let α ∈ Fi −mFi be a homogeneous element of degree Ti (M ).
Since F• is minimal and since Ti (M ) ≥ Ti+1 (M ), after possibly a change
of basis, the matrix representing ∂i+1 has a row of zeros. Now consider the
dual F•∗ := HomS (F• , S), whose maps are represented by the transposes of
the matrices representing the ∂i ; we denote them by ∂i∗ . Hence the matrix
representing ∂i+1∗ : Fi∗ → Fi+1
∗ has a column of zeros. In other words, there
is a basis element α (corresponding to the dual of α) of Fi∗ which is in the
∗

kernel of ∂i∗ . As F∗ was minimal, all entries of the matrix ∂i+1 , and hence of
∗ , are contained in the maximal ideal. Therefore α∗ is a nonzero element
∂i+1
∗ ) = Exti (M, S) and in particular Exti (M, S) 6= 0. This
in Ker(∂i∗ )/ Im(∂i+1 S S
is impossible since i < grade(M ) = codim(M ). 
It is also possible to prove the above result using the Boij-Söderberg de-
composition of the Betti table of M . (See [10] and [6].) This decomposition
involves multiples of Betti diagrams of pure Cohen-Macaulay S-modules of
codimension at least codim(M ). The associated degree sequences are strictly
increasing thus forcing the inequality in the above proposition.
Remark 2.3. We note that in general it is possible that ti (M ) ≥ ti+1 (M ) for
i ≥ codim(M ), even if M is cyclic. A Macaulay2 [18] computation shows
that if
S = K[x1 , x2 , y1 , y2 , z1 , z2 , z3 , z4 ]
and
I = x61 , y16 , x21 x42 + y12 y24 + x1 y1 (x32 z1 + x22 y2 z2 + x2 y22 z3 + y23 z4 ) ,

then T7 (S/I) = 38 while T8 (S/I) = 34. Note that codim(S/I) = 2.

One can also create more extreme examples by taking direct sums of
cyclic modules of different codimensions. For example, if S = K[x, y, z]
and M = S/(xn , y n ) ⊕ S/(x, y, z), then codim(M ) = 2, T2 (M ) = 2n and
T3 (M ) = 3.
An easy consequence of Theorem 1.3 and Proposition 2.2 is the following
statement regarding the regularity of Cohen-Macaulay modules.
Corollary 2.4. If M is a Cohen-Macaulay module of codimension c and
Ann(M ) contains a homogeneous Cohen-Macaulay ideal J also of codimen-
sion c, then
reg(M ) ≤ T0 (M ) + Tc (S/J) − c.
 ON THE MAXIMAL GRADED SHIFTS OF IDEALS AND MODULES 5

In particular, if Ann(M ) contains a regular sequence of degrees d1 , . . . , dc ,

then  
c
X
reg(M ) ≤ T0 (M ) +  dj  − c.
j=1

Proof. By Proposition 2.2, we have reg(M ) = Tc (M ) − c. Since pd(S/J) =

pd(M ) = c, the rest follows from Theorem 1.3. 
So if M is Cohen-Macaulay and Ann(M ) contains a regular sequence of
length codim(M ), then one needs only the maximal degree of a minimal
generator (i.e. the 0th step in the free resolution of M ) to get an effective
bound on reg(M ). In particular, the previous corollary shows that among
Cohen-Macaulay ideals of fixed codimension and maximal generating degree,
complete intersections have the maximum regularity. The goal of this paper
is to show that for arbitrary modules M , one need only compute the first
pd(M ) − codim(M ) steps of the resolution of M to get an effective bound
on the regularity of M .

3. Main Results
First we prove the previously mentioned special case of Conjecture 1.4.
The proof, which we include for completeness, is adapted from that of
Herzog-Srinivasan [14, Proposition 2]
Theorem 3.1. Let M be a graded S-module of projective dimension p.
Suppose f ∈ Ann(M ) and set d = deg(f ). Then
Tp (M ) ≤ Tp−1 (M ) + d = Tp−1 (M ) + T1 (S/(f )).
In particular, for any homogeneous ideal I of S, if p = pd(S/I) then
Tp (S/I) ≤ Tp−1 (S/I) + t1 (S/I).
Proof. Let F• denote the graded minimal free resolution of M with differen-
tial maps ∂i : Fi → Fi−1 . Let F•∗ = HomS (F• , S) be the dual complex with
differential maps ∂i∗ : Fi−1 ∗ → F ∗ . Let g , . . . , g denote a homogeneous basis
i 1 r
of Fp−1 and let h1 , . . . , hs denote a homogeneous basis of Fp . Let g1∗ , . . . , gr∗
and h∗1 , . . . , h∗s be the corresponding dual bases of Fp−1 ∗ and Fp∗ , respectively.
The map ∂p : Fp → Fp−1 is represented by a matrix (cij ) with respect to
these basis, and ∂p∗ is represented by (cij )T . Since Fp+1 ∗ = 0, ∂p+1∗ (h∗ ) = 0
i
∗
for all i. Since F• is minimal, for every i, hi is a nonzero minimal gener-
ator of H p (F•∗ ) = ExtpS (M, S). Since f ∈ Ann(M ) ⊆ Ann ExtpS (M, S) ,

f · h∗i ∈ Im(∂p∗ ) for all i.

Now fix i. Write ∂p (hi ) = rj=1 cij gj , and notice that cij 6= 0 for some j
P

since ∂p is injective. Note also that ∂p∗ (gj∗ ) = si=1 cij h∗i for all j. Therefore,
P

since f · h∗i ∈ Im(∂p∗ ),

r
X r
X s
X
f · h∗i = bj ∂p∗ (d∗j ) = bj cij h∗i ,
j=1 j=1 i=1
6 J. MCCULLOUGH

for some b1 , . . . , br ∈ S. Then since at least one of the cij 6= 0 we have

deg(f ) + deg(h∗i ) = deg(bj ) + deg(cij ) + deg(h∗i ),
and hence
d = deg(f ) = deg(bj ) + deg(cij ) ≥ deg(cij ).
Pr
Since ∂p (hi ) = j=1 cij gj we have
deg(hi ) = deg(cij ) + deg(gj ) ≤ d + Tp−1 (M )
for all i. Therefore Tp (M ) ≤ d + Tp−1 (M ).
The second statement follows by selecting f ∈ I of degree t1 (S/I). Then
since f ∈ Ann(S/I) we have
Tp (S/I) ≤ Tp−1 (S/I) + T1 (S/(f )) = Tp−1 (S/I) + t1 (S/I).

Note that t1 (S/I) is just the minimal degree of a minimal generator of
I. Theorem 3.1 recovers and strengthens [14, Corollary 3], [11, Theorem
2.1] and [17, Theorem 4.4]. The theorem implies the following bound on
regularity in terms of the first p − 1 syzygies:
Corollary 3.2. Let M be a graded S-module with pd(M ) = p. Suppose
f ∈ Ann(M ) and set d = deg(f ).
 
reg(M ) ≤ max max {Ti (M ) − i}, Tp−1 (M ) + d − p .
0≤i≤p−1

Proof. By definition reg(M ) ≤ max0≤i≤p {Ti (M ) − i}. By the previous the-

orem Tp (M ) − p ≤ Tp−1 + d − p. 
Next we prove a weaker bound on reg(M ) but one that is still only de-
pendent on the first pd(M ) − codim(M ) steps of the resolution. First we
make a notational definition for convenience. Let N denote P the set of non-
negative integers. Let a = (a1 , . . . , aq ) ∈ Nq . Define |a| = qi=1 ai (i + 1).
For instance |(1, 2, 0, 1)| = 1 · 2 + 2 · 3 + 0 · 4 + 1 · 5 = 13.
We now consider arbitrary modules M with a fixed Cohen-Macaulay ideal
J ⊆ Ann(M ). The assumption that J is Cohen-Macaulay is necessary to
make the argument work. In the case when J is a complete intersection of
linear forms, then S = S/J is a polynomial ring and the regularity of M
may be computed over S. Hence we focus on the case when J has at least
one generator of at least degree 2. Our main result is the following:
Theorem 3.3. Let M be a finitely generated S-module. Suppose that Ann(M )
contains a homogeneous ideal J such that S/J is Cohen-Macaulay with
c = codim(J) = pd(S/J) and T1 (S/J) ≥ 2. Set p = pd(M ). Then
 
 Xc 
reg(M ) ≤ max Ti (M ) + aj (Tj (S/J)) + Tc (S/J) − p.
0≤i≤p−c  
c
a∈N j=1
|a|≤p−c−i
 ON THE MAXIMAL GRADED SHIFTS OF IDEALS AND MODULES 7

Proof. Set R = S/J. Since M is an R-module, p = pdS (M ) ≥ codim(M ) ≥

codim(R) = c. Note that if c = 1, then the statement follows from Corol-
lary 3.2.
We proceed by induction on p.
If p = c, then M is Cohen-Macaulay. Hence reg(M ) ≤ T0 (M )+Tc (S/J)−c
by Corollary 2.4, which proves the base case, since 0 = (0, 0, . . . , 0) ∈ Nc is
the only vector a ∈ Nc such that |a| ≤ 0.
If p > c, then consider the short exact sequence of R-modules (and thus
also S-modules)
0 → K → F → M → 0.

Here F = ⊕j R(−dj )β0,j (M ) is the 0th step in a minimal R-free resolution of

M , and K = Ker(F ։ M ). (In other words, K is the module of first syzygies
of M over R.) Since F is free over R, an S-free resolution of F is merely an
S-free resolution of S/J tensored by ⊕j S(−dj )β0,j (M ) over S. In particular,
pdS (F ) = pdS (R) = c < p = pdS (M ) and Ti (F ) = T0 (F ) + Ti (S/J) for
0 ≤ i ≤ c. Moreover, by our choice of F we have T0 (F ) = T0 (M ). It follows
that pdS (K) = pdS (M ) − 1 = p − 1. Since pdS (S/J) = c < p = pdS (M ),
Tp (M ) ≤ Tp−1 (K).
If p = c + 1, then dim(M ) − depth(M ) = pd(M ) − codim(M ) = 1. The
only vector a ∈ Nc with |a| ≤ 1 is again 0. Again by Corollary 2.4, we
have Tp (M ) ≤ T1 (M ) + Tp−1 (S/J). It follows from Proposition 2.2 that
reg(M ) = max{Tp (M ) − p, Tp−1 (M ) − (p − 1)}. Thus it suffices to show that
Tp−1 (M ) ≤ T1 (M ) + Tp−1 (S/J) − 1. This follows by noting

Tp−1 (M ) ≤ max{Tp−1 (F ), Tp−2 (K)}

≤ max{T0 (F ) + Tp−1 (S/J), Tp−1 (K) − 1}
≤ max{T0 (M ) + Tp−1 (S/J), T0 (K) + Tp−1 (S/J) − 1}
≤ T1 (M ) + Tp−1 (S/J) − 1.

The first inequality is just Lemma 2.1; the second inequality follows from
the discussion of Ti (F ) and from Proposition 2.2; the third inequality follows
from Theorem 1.3 applied to K; the final inequality is again Proposition 2.2
and Lemma 2.1.
If p > c+1, then since J ⊂ Ann(K) and since pd(K) < p, by the inductive
hypothesis we have
c
X
reg(K) ≤ Ti (K) + aj Tj (S/J) + Tc (S/J) − (p − 1),
j=1

for some 0 ≤ i ≤ (p − 1) − c and some a ∈ Nc with |a| ≤ (p − 1) − c − i.

Using the short exact sequence above we have

reg(M ) ≤ max{reg(F ), reg(K) − 1}.

8 J. MCCULLOUGH

Since F is Cohen-Macaulay,

reg(F ) = Tc (F ) − c
= T0 (F ) + Tc (S/J) − c
= T0 (M ) + Tc (S/J) − c.

If p − c is even, say p − c = 2m for some nonnegative integer m, then we set

a = (m, 0, 0, . . .) and note that |a| = 2m = p−c. If p−c is odd, since p > c+1,
p must be at least 3. Hence we can write p−c = 2m+3 for some nonnegative
integer m. Noting that c ≥ 2 by assumption, we set a = (m, 1, 0, 0, . . .) and
see that again |a| = 2m + 3 =Pp − c. Since T1 (S/J)Pc ≥ 2 we have Tj (S/J) ≥
c
j+1 for 1 ≤ j ≤ c. Therefore j=1 aj Tj (S/J) ≥ j=1 aj (j+1) = |a| = p−c;
thus, in either case we have
c
X
reg(F ) ≤ T0 (M ) + aj Tj (S/J) + Tc (S/J) − p.
j=1

Hence we may assume that

c
X
reg(M ) ≤ reg(K) − 1 = Ti (K) + aj Tj (S/J) + Tc (S/J) − p,
j=1

for some 0 ≤ i ≤ c and some a ∈ Nc with |a| ≤ (p−1)−c−i. By Lemma 2.1,

Ti (K) ≤ Ti+1 (M ) or Ti (K) ≤ Ti (F ). In the former case we have
c
X
reg(M ) ≤ Ti+1 (M ) + aj Tj (S/J) + Tc (S/J) − p.
j=1

In the latter case, since Ti (F ) = T0 (M ) + Ti (S/J), we have

c
X
reg(M ) ≤ T0 (M ) + Ti (S/J) + aj Tj (S/J) + Tc (S/J) − p
j=1
c
X
= T0 (M ) + a′j Tj (S/J) + Tc (S/J) − p,
j=1

where
(
aj if j 6= i
a′j =
aj + 1 if j = i.
 ON THE MAXIMAL GRADED SHIFTS OF IDEALS AND MODULES 9

Finaly we note that

c
X
′
|a | = a′j (j + 1)
j=1
c
X
= (i + 1) + aj (j + 1)
j=1
= (i + 1) + |a|
≤ (i + 1) + (p − 1) − c − i
=p−c
as desired. 
As stated, the bound may be a bit hard to interpret. We include a specific
instance of its application below.
Example 3.4. Suppose pd(M ) = 7 and codim(M ) = 3. Further suppose
that J = (x, y, z)2 ⊂ Ann(M ). Then Ti (S/J) = i + 1 for 1 ≤ i ≤ 3. We
consider vectors a ∈ N3 with |a| = p − c ≤ 4. No vectors satisfy |a| = 1; only
the vectors (1, 0, 0) and (0, 1, 0) satisfy |a| = 2 and 3, respectively. The only
such vectors a with |a| = 4 are (2, 0, 0) and (0, 0, 1). Therefore, the previous
theorem implies that
reg(M ) ≤ max{T4 (M ) + T3 (S/J) − 7, T2 (M ) + T1 (S/J) + T3 (S/J) − 7,
T1 (M ) + T2 (S/J) + T3 (S/J) − 7, T0 (M ) + 2T3 (S/J) − 7,
T0 (M ) + 2T1 (S/J) + T3 (S/J) − 7}
= max{T4 (M ) − 3, T2 (M ) − 1, T1 (M ), T0 (M ) + 1}
= max{T4 (M ) − 3, T2 (M ) − 1},
where the second equality follows since codim(M ) = 3 and hence T0 (M ) <
T1 (M ) < T2 (M ) < T3 (M ) by Proposition 2.2.
As a corollary to our main theorem, we get a weak upper bound on Tp (M )
in terms of the same data.
Corollary 3.5. Let M be a finitely generated S-module. Suppose that
Ann(M ) contains a homogeneous ideal J such that S/J is Cohen-Macaulay
with c = codim(J) = pd(S/J). Set p = pd(M ). Then
 
 c
X 
Tp (M ) ≤ max Ti (M ) + aj Tj (S/J) + Tc (S/J).
0≤i≤p−c  
c
a∈N j=1
|a|≤p−c−i

Proof. Since Tp (M ) ≤ reg(M ) + p, the inequality follows from Theorem 3.3.


Finally we state a few special cases of the main theorem in the cyclic case.
10 J. MCCULLOUGH

Corollary 3.6. Let I be an ideal of S with p = pd(S/I) and c = codim(I).

If I contains a regular sequence of forms of common degree d then
reg(S/I) ≤ max {Ti (S/I) + (p − i) d} − p.
0≤i≤p−c

Proof. Let f1 , . . . , fc be a regular sequence of forms of degree d contained in

I. Then J = (f1 , . . . , fc ) ⊂ Ann(S/I) is a Cohen-Macualay ideal and, since
S/J is resolved by a Koszul complex on f1 , . . . , fc , we have Ti (S/J) = di for
0 ≤ i ≤ c. By Theorem 3.3 we have for some 0 ≤ i ≤ p − c and some a ∈ Nc
with |a| ≤ p − c − i that
c
X
reg(S/I) ≤ Ti (S/I) + aj (Tj (S/J)) + Tc (S/J) − p
j=1
c
X
= Ti (S/I) + (aj dj) + dc − p
j=1
 
c
X
≤ Ti (S/I) + d  (aj (j + 1)) + dc − p
j=1

≤ Ti (S/I) + d(p − c − i) + dc − p
= Ti (S/I) + d(p − i) − p

Thus we get a bound on the regularity of any cyclic module S/I purely
in terms of the first pd(S/I) − codim(I) maximal graded shifts:
Corollary 3.7. Let I be an ideal of S with p = pd(S/I) and c = codim(I).
Then
reg(S/I) ≤ max {Ti (S/I) + (p − i) T1 (S/I)} − p.
0≤i≤p−c

Proof. This follows from the previous corollary since I contains a regular
sequence of forms of length c and of degree at most T1 (S/I). 

This upper bound is likely not tight for all values of p and c, but it seems
close to a best possible result in the following sense:
Fix a value of c. A rephrasing of the Ullery’s result implies that for p ≫ 0
there is no bound on reg(S/I) purely in terms of T1 (S/I), . . . , Tp−c−1 (S/I),
where p = pd(S/I) and c = codim(I). Thus the result above, which gives
a linear bound on reg(S/I) in terms of T1 (S/I), . . . , Tp−c (S/I) and pd(S/I)
uses minimal data in some sense. Where Ullery’s result shows that almost
any sequences of increasing positive integers appear as the initial maximal
graded shifts of some cyclic module or ideal, our result restricts what can
happen at the end of the resolution.
 ON THE MAXIMAL GRADED SHIFTS OF IDEALS AND MODULES 11

Acknowledgements
The author thanks Lance Miller, Irena Peeva and Hema Srinivasan for
useful feedback on an earlier draft of this paper.

References
[1] L. Avramov, A. Conca and S. Iyengar, Free resolutions over commutative Koszul
algebras, Math. Res. Lett. 17 (2010), no. 2, 197–210.
[2] L. Avramov, A. Conca and S. Iyengar, Subadditivity of syzygies of Koszul algebras.
Math. Ann. 361 (2015), no. 1-2, 511–534.
[3] J. Beder, J. McCullough, L. Núnez-Betancourt, Luis, A. Seceleanu, B. Snapp, and
B. Stone, Ideals with larger projective dimension and regularity, J. Symbolic Com-
put. 46 (2011), no. 10, 1105-1113.
[4] D. Bayer and D. Mumford: What can be computed in Algebraic Geometry?, Compu-
tational Algebraic Geometry and Commutative Algebra, Symposia Mathematica,
Volume XXXIV, Cambridge University Press, Cambridge, 1993, 1–48.
[5] D. Bayer and M. Stillman: On the complexity of computing syzygies. Computational
aspects of commutative algebra, J. Symbolic Comput. 6 (1988), 135–147.
[6] M. Boij and J. Söderberg, Betti numbers of graded modules and the multiplicity
conjecture in the non-Cohen-Macaulay case, Algebra Number Theory 6 (2012), no.
3, 437–454.
[7] G. Caviglia and E. Sbarra: Characteristic-free bounds for the Castelnuovo-Mumford
regularity, Compos. Math. 141 (2005), 1365–1373.
[8] D. Eisenbud, Commutative algebra. With a view toward algebraic geometry. Grad-
uate Texts in Mathematics, 150. Springer-Verlag, New York (1995).
[9] D. Eisenbud, C. Huneke and B. Ulrich, The regularity of Tor and graded Betti
numbers, Amer. J. Math. 128 (2006), 573–605.
[10] D. Eisenbud and F.-O. Schreyer, Betti numbers of graded modules and cohomology
of vector bundles, J. Amer. Math. Soc. 22 (2009), no. 3, 859–888.
[11] S. El Khouri and H. Srinivasan, A note on the subadditivity of syzygies, preprint:
arXiv:1602.02116.
[12] M. Green and R. Lazarsfeld, Some results on the syzygies of finite sets and algebraic
curves, Compositio Math. 67 (1988), no. 3, 301–314.
[13] J. Harris, Algebraic geometry. A first course. Graduate Texts in Mathematics 133.
Springer-Verlag, New York (1992).
[14] J. Herzog and H. Srinivasan, On the subadditivity problem for maximal shifts in free
resolutions, Commutative algebra and noncommutative algebraic geometry. Vol. II,
Math. Sci. Res. Inst. Publ. 68, Cambridge Univ. Press, New York (2015), 245–249.
[15] J. Koh: Ideals generated by quadrics exhibiting double exponential degrees, J. Alge-
bra 200 (1998), 225–245.
[16] E. Mayr and A. Meyer: The complexity of the word problem for commutative semi-
groups and polynomial ideals, Adv. in Math. 46 (1982), 305–329.
[17] J. McCullough, A polynomial bound on the regularity of an ideal in terms of half of
the syzygies, Math. Res. Lett. 19 (2012), no. 3, 555–565.
[18] Macaulay 2, a software system for research in algebraic geometry. Available at
http://www.math.uiuc.edu/Macaulay2/
[19] B. Ullery, Designer ideals with high Castelnuovo-Mumford regularity, Math. Res.
Lett. 21 (2014), no. 5, 1215–1225.

Department of Mathematics, Iowa State University, Ames, IA 50011

E-mail address: jmccullo@iastate.edu